Hi there ! If you are looking for the code and a live demo app , here's a similar blog post which would describe the whole idea better : Link - medium.com/@avra42/build-a-personal-search-engine-web-app-using-open-ai-text-embeddings-d6541f32892d
Hey this is so kind of you . Apologies for my late response. Please leave your email ID , I'm happy to share the crude notebook script with you ( incase you want it right away )
Really amazing video bro :) Thanks for making such amazing content :) Quick question. Can you make a video for a Streamlit app where you teach us how can we assign login credentials to our app? :)
Hey thanks for your kind words, means a lot . Defnitely, this is pretty common requirement I see over many forums. However, have you tried this alternative way - using Firebase ? I've a video on it - old one, but seems to be working - video link : Part 1 - th-cam.com/video/KLmSfHcOXlc/w-d-xo.html Part 2 - th-cam.com/video/uzc3OEjjbn8/w-d-xo.html
Yes - because of the search term embeddings it would cost . However it’s pretty less - 0.004 /1k token ( if I’m not wrong ) - text-Ada is one of cheapest available api with pretty fast performance I would say
I have one related question on Embeddings. So every time a user asks the bot a question, the bot searches the knowledgebase and answers. the following use case impossible: USER: How can I register for starter training ? BOT: If you meet all the conditions, you can register from the link … USER: What does it cost ? BOT: I dont know. how can i deal with this problem.Since the user specified the course name in the previous question, but not in this question, there will be no match from the knowledgebase. How can I search while keeping the context from the previous question?
Haha I know exactly what you are looking for - I’ve another tutorial which exactly explains how to create a bot which has memory - How to Build a CHAT BOT with ChatGPT API (GPT-3.5-TURBO) having CONVERSATIONAL MEMORY in Python th-cam.com/video/cHjlperESbg/w-d-xo.html A blog post - medium.com/@avra42/how-to-build-a-chatbot-with-chatgpt-api-and-a-conversational-memory-in-python-8d856cda4542 Let me know if it solves your use case . Try the demo app as well for a quick overview .
@@Avra_b Actually, problem is not with the chatbot's memory, it already has a memory, it remembers the conversation history. When the user asks a question, bot first looks at the embedding (document), if there is a related document answers using the context in the document. If there is no related embedding (document), chatgpt itself responds directly. The problem is that when the question "what does it cost" is asked, it can't match the embeddings because it has a low similarity score and can't do it while actually expected to respond by looking at the embedding (document). If the question were "What does starter training", the associated embedding would be found and the correct answer could be generated within that context. I hope I was able to explain the problem, if it is not understood, I can make an additional explanation.
Thanks for your query . Hard to understand. Your previous example was more about Memory issue - since the context is not stored . Your present example is not clear to me. But I would say usually with such Q&A the knowledge base works pretty well . Try LangChain ( I’ve videos on it ) . Best wishes
Haha NO ! Even if you deploy your own app over Streamlit cloud - the secrets are well protected . You can read more about it in @streamlitofficial cloud documentation about their secrets security protocol . Cheers
Hi there ! If you are looking for the code and a live demo app , here's a similar blog post which would describe the whole idea better :
Link - medium.com/@avra42/build-a-personal-search-engine-web-app-using-open-ai-text-embeddings-d6541f32892d
Thanks!
Hey this is so kind of you . Apologies for my late response. Please leave your email ID , I'm happy to share the crude notebook script with you ( incase you want it right away )
Thank you for creating such a nice video with end-to-end flow.
Hey means a lot . Thank you 🤗 stay tuned :)
A much needed video thank you very much 🥳🥳🔥
I’m glad that you liked it . Always happy to hear from you :)
Can we also download the relevant documents (files)from the results?
Yeah . Download via st.download functionality .
Thanks 👍
can we use different LLM with different Embedding model
Certainly ! I've videos with Cohere LLMs. Check those videos out . let me know how it goes :)
Thanks a lot will do
This video is detailed AF. Thank you
Could you please also share the Google Colab notebooks link?
Hey Manoj, glad to hear that you liked this video.
Please give me few more days, you will find the notebook as well as the tutorial in blog post form.
Really amazing video bro :) Thanks for making such amazing content :)
Quick question. Can you make a video for a Streamlit app where you teach us how can we assign login credentials to our app? :)
Hey thanks for your kind words, means a lot . Defnitely, this is pretty common requirement I see over many forums. However, have you tried this alternative way - using Firebase ? I've a video on it - old one, but seems to be working - video link :
Part 1 - th-cam.com/video/KLmSfHcOXlc/w-d-xo.html
Part 2 - th-cam.com/video/uzc3OEjjbn8/w-d-xo.html
Hey bro, what camera are you using?
Hi ! Usually I use my iphone as my recording camera :D
@@Avra_b That's good. Good quality 🙌🙌
Thank you.
Welcome 🤗
Awesome
Glad that you liked it :)
does it cost money every time when user does a search?
Yes - because of the search term embeddings it would cost . However it’s pretty less - 0.004 /1k token ( if I’m not wrong ) - text-Ada is one of cheapest available api with pretty fast performance I would say
I have one related question on Embeddings. So every time a user asks the bot a question, the bot searches the knowledgebase and answers. the following use case impossible:
USER: How can I register for starter training ?
BOT: If you meet all the conditions, you can register from the link …
USER: What does it cost ?
BOT: I dont know.
how can i deal with this problem.Since the user specified the course name in the previous question, but not in this question, there will be no match from the knowledgebase. How can I search while keeping the context from the previous question?
Haha I know exactly what you are looking for - I’ve another tutorial which exactly explains how to create a bot which has memory - How to Build a CHAT BOT with ChatGPT API (GPT-3.5-TURBO) having CONVERSATIONAL MEMORY in Python
th-cam.com/video/cHjlperESbg/w-d-xo.html
A blog post - medium.com/@avra42/how-to-build-a-chatbot-with-chatgpt-api-and-a-conversational-memory-in-python-8d856cda4542
Let me know if it solves your use case . Try the demo app as well for a quick overview .
@@Avra_b Actually, problem is not with the chatbot's memory, it already has a memory, it remembers the conversation history. When the user asks a question, bot first looks at the embedding (document), if there is a related document answers using the context in the document. If there is no related embedding (document), chatgpt itself responds directly. The problem is that when the question "what does it cost" is asked, it can't match the embeddings because it has a low similarity score and can't do it while actually expected to respond by looking at the embedding (document). If the question were "What does starter training", the associated embedding would be found and the correct answer could be generated within that context. I hope I was able to explain the problem, if it is not understood, I can make an additional explanation.
Thanks for your query . Hard to understand. Your previous example was more about Memory issue - since the context is not stored . Your present example is not clear to me. But I would say usually with such Q&A the knowledge base works pretty well . Try LangChain ( I’ve videos on it ) . Best wishes
just embed the entire conversation each time, or have it at least embed the last "n" question answers.
Nice way to get people's openAI API keys for free hahaha
But how ? Curious 🧐
@@Avra_b my bad I thought your app was publicly accessible and that people would enter (api key + request) to use it
Haha NO ! Even if you deploy your own app over Streamlit cloud - the secrets are well protected . You can read more about it in @streamlitofficial cloud documentation about their secrets security protocol . Cheers